Update README.md
Browse files
README.md
CHANGED
|
@@ -1,12 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
| 1 |
# superhot-30b-8k-4bit-128g-safetensors
|
| 2 |
|
|
|
|
|
|
|
| 3 |
Merged base LLaMA and LoRA with this:
|
| 4 |
https://github.com/tloen/alpaca-lora
|
| 5 |
|
| 6 |
Base LLaMA 30B:
|
| 7 |
https://huggingface.co/huggyllama/llama-30b
|
| 8 |
|
| 9 |
-
|
| 10 |
https://huggingface.co/kaiokendev/superhot-30b-8k-no-rlhf-test
|
| 11 |
|
| 12 |
``` sh
|
|
@@ -56,4 +61,4 @@ CUDA_VISIBLE_DEVICES=0 python test_benchmark_inference.py \
|
|
| 56 |
-- Loading dataset...
|
| 57 |
-- Testing 40 chunks....
|
| 58 |
** Perplexity: 4.6612
|
| 59 |
-
```
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: other
|
| 3 |
+
---
|
| 4 |
# superhot-30b-8k-4bit-128g-safetensors
|
| 5 |
|
| 6 |
+
**Note: Maximum sequence length (max_seq_len) and compression factor (compress_pos_emb) need to be set to 8192 and 4.**
|
| 7 |
+
|
| 8 |
Merged base LLaMA and LoRA with this:
|
| 9 |
https://github.com/tloen/alpaca-lora
|
| 10 |
|
| 11 |
Base LLaMA 30B:
|
| 12 |
https://huggingface.co/huggyllama/llama-30b
|
| 13 |
|
| 14 |
+
SuperHOT 30B 8k no-rlhf-test LoRA:
|
| 15 |
https://huggingface.co/kaiokendev/superhot-30b-8k-no-rlhf-test
|
| 16 |
|
| 17 |
``` sh
|
|
|
|
| 61 |
-- Loading dataset...
|
| 62 |
-- Testing 40 chunks....
|
| 63 |
** Perplexity: 4.6612
|
| 64 |
+
```
|